Available in June, 2026
High-performance computing (HPC) and AI are increasingly converging around accelerated, heterogeneous processors with each type of processor doing what it does best. For example, the latest platforms combine CPUs with GPUs, application-specific accelerators, high-bandwidth memory and advanced interconnects. The rapid growth of AI has made accelerators the principal source of industry expansion, while scientific computing, simulation, national labs and weather forecasting continue to support demand for conventional HPC systems. Meanwhile, the HPC and AI processor markets has become less distinct.
This report analyses both the technology and demand changes to better understand where the money will be made in the future. It also examines the structure of this industry sector and how it might change in the future. For now, NVIDIA is the dominant supplier of data-center AI processors, while AMD is NVIDIA’s most important merchant-market challenger. CPUs are also a factor in this market and here we note that intel remains a major supplier of server CPUs used in HPC clusters and AI infrastructure. Other influencers in the HPC/AI process space include IBM, Fujitsu, Huawei, Cerebras, Groq, Tenstorrent and SambaNova. Strategic profiles of these companies are provided in the report covering product development, market strategy and manufacturing.
The industry structure is also being reshaped by custom silicon developed by hyperscale cloud companies. Google designs TPUs, Amazon Web Services offers Trainium and Inferentia processors, and Microsoft has developed its Maia accelerators. These companies are both major purchasers of merchant processors and potential competitors to their suppliers. We discuss the implications of this in the report as well as explaining what it means that hyperscalers generally provide their processors as part of cloud services rather than as standalone components.
Competition extends well beyond processor specifications. Success in this market increasingly depends on software tools, optimized libraries, memory capacity, chip-to-chip interconnects, networking, power efficiency and the ability to deliver complete systems at scale. The industry is also constrained by access to leading-edge semiconductor fabrication, advanced packaging and high-bandwidth memory, creating significant dependence on a small number of foundries and component manufacturers. All of this is covered in this new CIR report.
